Optimization of therapeutic drugs for safety and efficacy is a delicate balancing act requiring chemical mod- ification of the drug candidate without altering it's therapeutic benefits. Knowledge of a drug's mechanism of action is critical for successful optimization. In prior work, we developed a novel approach for predict- ing drug mechanisms from gene expression profiles of drug activity. The method was pioneering in its use of a mathematical gene network model to analyze noisy, high-dimensional gene expression data. It was successfully applied in a preliminary study on yeast, but certain characteristics of our early work limit its accuracy and broader applicability. In particular, the modeling approach obscured biologi- cally meaningful network relationships, and could not incorporate prior data on gene network structure to improve predictions. In this project, we will develop statistical, computational, and experimental methods to extend and broaden our previous work on predicting drug mechanism of action from gene expression data. In aims 1-3, we will adapt the framework of simultaneous equation models (SEMs) to our problem and, within this context, develop extensions of recent techniques for sparse inference, both frequentist and Bayesian variants. In aims 4-6, we will test and validate the sparse SEM methods on a database of expression profiles that we will obtain by treating yeast with a large set of stresses and bioactive compounds. The resulting methods will enable rapid and inexpensive discovery of the mechanism of action of both candidate therapeutic compounds and hazardous biological toxins. This proposal includes innovative contributions in biology, bioinformatics, statistics, chemical biology and pharmacology. The work will make valuable contributions to the fields of systems biology and bioinformatics by demonstrating unique methodology for design and analysis of microarray experiments and the practical utility of genome-scale modeling of gene regulation. To the statistics community, this work will offer powerful extensions of the SEM framework, through its pairing with adaptive, sparse inference tools. For the chemical biology and pharmacology communities, this work will offer a valuable new method to determine the biological activity of novel compound classes. In addition, this work could substantially accelerate the development of safe and effective therapeutic drugs. It could provide a novel computational tool for quickly and cost-effectively evaluating the mechanism of action of chemical and biological agents of potential therapeutic value. Moreover, some of the compounds that we propose to study in this project are novel anticancer and antifungal compounds with therapeutic potential.